On 11/22/2017 9:49 AM, Greg KH wrote:

quoted

quoted

If you have out-of-tree drivers, then yes, they can make a wrapper for
this function like this if they really feel the need, or they can get
their code merged:)

Sorry, I guess I should have been clearer.  My suggestion was to fix some of
the drivers where the domain can be determined, and for the rest, just mark
the old function as deprecated.

So the build now gets warnings?  That's annoying, and then someone else
will have to make the exact same patches that were created here?

quoted

If that's still a terrible idea, well, okay.  I'm just unsure that simply
hard-coding a 0 for the domain for some drivers is really a solution.  Don't
we really want all drivers to properly support all domains?

I bet all of those drivers don't care because they are running only in
systems with 1 domain, otherwise they would be broken today, right?  But
really, it shouldn't be that hard to get to the "real" PCI device to
provide the correct pointer to the domain for most of these, as I
pointed out in one patch review already.

I agree. Point is 95% drivers do support multiple segments. These are the
exceptions. 

We should try to fix them as much as we can. (I'll take a look at Greg's
suggestion)

In the end, this API is a backdoor and workarounds the proper APIs and
usual practices like carrying struct pci_dev pointer.

I'm trying to shoot that API in the foot so that a developer thinks twice
before putting number 0 there.

-- 
Sinan Kaya
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.